A Reliable Network on Chip based using Whale Firefly Optimization and Deep Neural Network

Abstract

Network on Chip (NoC) provides the technology for forming interconnect pattern to perform a task. NoC suffers from high power consumption and low reliability to perform computation. Prediction of network congestion helps to effectively handle the network parameters and increases the network performance. Existing reinforcement learning and decision tree methods have the limitation of overfitting problem and unstable performance. In this research, Whale Firefly Optimization (WFO) of hybrid optimization method with Deep Neural Network (DNN) is proposed to improve the reliability and reduces the power consumption in NoC. The WFO-DNN method tends to provides the optimized reward value based on the latency, power and traffic congestion. Traffic congestion is measured from the simulated network and applied for the WFO-DNN method to provide optimal reward update. The hybrid optimization method is applied for parameter settings for the NoC to maintain the trade-off for latency, power and reliability. The hybrid method of WFO have advantages of good convergence and improve the exploration of the method. The DNN model provides the efficient performance in the prediction of faults in the network to improve the reliability. The DNN model is tested with various number of hidden layer to find the suitable number of layer. The proposed WFO-DNN model has 1.34 improve of speed, 1.94 energy efficiency and existing reinforcement method has 1.24 improvement of speed, and 1.77 energy efficiency.